Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
[deliverable/linux.git] / include / linux / blk-cgroup.h
1 #ifndef _BLK_CGROUP_H
2 #define _BLK_CGROUP_H
3 /*
4 * Common Block IO controller cgroup interface
5 *
6 * Based on ideas and code from CFQ, CFS and BFQ:
7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
8 *
9 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
10 * Paolo Valente <paolo.valente@unimore.it>
11 *
12 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
13 * Nauman Rafique <nauman@google.com>
14 */
15
16 #include <linux/cgroup.h>
17 #include <linux/u64_stats_sync.h>
18 #include <linux/seq_file.h>
19 #include <linux/radix-tree.h>
20 #include <linux/blkdev.h>
21 #include <linux/atomic.h>
22
23 /* Max limits for throttle policy */
24 #define THROTL_IOPS_MAX UINT_MAX
25
26 #ifdef CONFIG_BLK_CGROUP
27
28 enum blkg_rwstat_type {
29 BLKG_RWSTAT_READ,
30 BLKG_RWSTAT_WRITE,
31 BLKG_RWSTAT_SYNC,
32 BLKG_RWSTAT_ASYNC,
33
34 BLKG_RWSTAT_NR,
35 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
36 };
37
38 struct blkcg_gq;
39
40 struct blkcg {
41 struct cgroup_subsys_state css;
42 spinlock_t lock;
43
44 struct radix_tree_root blkg_tree;
45 struct blkcg_gq *blkg_hint;
46 struct hlist_head blkg_list;
47
48 struct blkcg_policy_data *pd[BLKCG_MAX_POLS];
49
50 struct list_head all_blkcgs_node;
51 #ifdef CONFIG_CGROUP_WRITEBACK
52 struct list_head cgwb_list;
53 #endif
54 };
55
56 struct blkg_stat {
57 struct u64_stats_sync syncp;
58 uint64_t cnt;
59 };
60
61 struct blkg_rwstat {
62 struct u64_stats_sync syncp;
63 uint64_t cnt[BLKG_RWSTAT_NR];
64 };
65
66 /*
67 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
68 * request_queue (q). This is used by blkcg policies which need to track
69 * information per blkcg - q pair.
70 *
71 * There can be multiple active blkcg policies and each has its private
72 * data on each blkg, the size of which is determined by
73 * blkcg_policy->pd_size. blkcg core allocates and frees such areas
74 * together with blkg and invokes pd_init/exit_fn() methods.
75 *
76 * Such private data must embed struct blkg_policy_data (pd) at the
77 * beginning and pd_size can't be smaller than pd.
78 */
79 struct blkg_policy_data {
80 /* the blkg and policy id this per-policy data belongs to */
81 struct blkcg_gq *blkg;
82 int plid;
83
84 /* used during policy activation */
85 struct list_head alloc_node;
86 };
87
88 /*
89 * Policies that need to keep per-blkcg data which is independent
90 * from any request_queue associated to it must specify its size
91 * with the cpd_size field of the blkcg_policy structure and
92 * embed a blkcg_policy_data in it. cpd_init() is invoked to let
93 * each policy handle per-blkcg data.
94 */
95 struct blkcg_policy_data {
96 /* the policy id this per-policy data belongs to */
97 int plid;
98 };
99
100 /* association between a blk cgroup and a request queue */
101 struct blkcg_gq {
102 /* Pointer to the associated request_queue */
103 struct request_queue *q;
104 struct list_head q_node;
105 struct hlist_node blkcg_node;
106 struct blkcg *blkcg;
107
108 /*
109 * Each blkg gets congested separately and the congestion state is
110 * propagated to the matching bdi_writeback_congested.
111 */
112 struct bdi_writeback_congested *wb_congested;
113
114 /* all non-root blkcg_gq's are guaranteed to have access to parent */
115 struct blkcg_gq *parent;
116
117 /* request allocation list for this blkcg-q pair */
118 struct request_list rl;
119
120 /* reference count */
121 atomic_t refcnt;
122
123 /* is this blkg online? protected by both blkcg and q locks */
124 bool online;
125
126 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
127
128 struct rcu_head rcu_head;
129 };
130
131 typedef void (blkcg_pol_init_cpd_fn)(const struct blkcg *blkcg);
132 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
133 typedef void (blkcg_pol_online_pd_fn)(struct blkcg_gq *blkg);
134 typedef void (blkcg_pol_offline_pd_fn)(struct blkcg_gq *blkg);
135 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
136 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
137
138 struct blkcg_policy {
139 int plid;
140 /* policy specific private data size */
141 size_t pd_size;
142 /* policy specific per-blkcg data size */
143 size_t cpd_size;
144 /* cgroup files for the policy */
145 struct cftype *cftypes;
146
147 /* operations */
148 blkcg_pol_init_cpd_fn *cpd_init_fn;
149 blkcg_pol_init_pd_fn *pd_init_fn;
150 blkcg_pol_online_pd_fn *pd_online_fn;
151 blkcg_pol_offline_pd_fn *pd_offline_fn;
152 blkcg_pol_exit_pd_fn *pd_exit_fn;
153 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
154 };
155
156 extern struct blkcg blkcg_root;
157 extern struct cgroup_subsys_state * const blkcg_root_css;
158
159 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
160 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
161 struct request_queue *q);
162 int blkcg_init_queue(struct request_queue *q);
163 void blkcg_drain_queue(struct request_queue *q);
164 void blkcg_exit_queue(struct request_queue *q);
165
166 /* Blkio controller policy registration */
167 int blkcg_policy_register(struct blkcg_policy *pol);
168 void blkcg_policy_unregister(struct blkcg_policy *pol);
169 int blkcg_activate_policy(struct request_queue *q,
170 const struct blkcg_policy *pol);
171 void blkcg_deactivate_policy(struct request_queue *q,
172 const struct blkcg_policy *pol);
173
174 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
175 u64 (*prfill)(struct seq_file *,
176 struct blkg_policy_data *, int),
177 const struct blkcg_policy *pol, int data,
178 bool show_total);
179 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
180 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
181 const struct blkg_rwstat *rwstat);
182 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
183 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
184 int off);
185
186 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
187 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
188 int off);
189
190 struct blkg_conf_ctx {
191 struct gendisk *disk;
192 struct blkcg_gq *blkg;
193 u64 v;
194 };
195
196 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
197 const char *input, struct blkg_conf_ctx *ctx);
198 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
199
200
201 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
202 {
203 return css ? container_of(css, struct blkcg, css) : NULL;
204 }
205
206 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
207 {
208 return css_to_blkcg(task_css(tsk, blkio_cgrp_id));
209 }
210
211 static inline struct blkcg *bio_blkcg(struct bio *bio)
212 {
213 if (bio && bio->bi_css)
214 return css_to_blkcg(bio->bi_css);
215 return task_blkcg(current);
216 }
217
218 static inline struct cgroup_subsys_state *
219 task_get_blkcg_css(struct task_struct *task)
220 {
221 return task_get_css(task, blkio_cgrp_id);
222 }
223
224 /**
225 * blkcg_parent - get the parent of a blkcg
226 * @blkcg: blkcg of interest
227 *
228 * Return the parent blkcg of @blkcg. Can be called anytime.
229 */
230 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
231 {
232 return css_to_blkcg(blkcg->css.parent);
233 }
234
235 /**
236 * blkg_to_pdata - get policy private data
237 * @blkg: blkg of interest
238 * @pol: policy of interest
239 *
240 * Return pointer to private data associated with the @blkg-@pol pair.
241 */
242 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
243 struct blkcg_policy *pol)
244 {
245 return blkg ? blkg->pd[pol->plid] : NULL;
246 }
247
248 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
249 struct blkcg_policy *pol)
250 {
251 return blkcg ? blkcg->pd[pol->plid] : NULL;
252 }
253
254 /**
255 * pdata_to_blkg - get blkg associated with policy private data
256 * @pd: policy private data of interest
257 *
258 * @pd is policy private data. Determine the blkg it's associated with.
259 */
260 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
261 {
262 return pd ? pd->blkg : NULL;
263 }
264
265 /**
266 * blkg_path - format cgroup path of blkg
267 * @blkg: blkg of interest
268 * @buf: target buffer
269 * @buflen: target buffer length
270 *
271 * Format the path of the cgroup of @blkg into @buf.
272 */
273 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
274 {
275 char *p;
276
277 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
278 if (!p) {
279 strncpy(buf, "<unavailable>", buflen);
280 return -ENAMETOOLONG;
281 }
282
283 memmove(buf, p, buf + buflen - p);
284 return 0;
285 }
286
287 /**
288 * blkg_get - get a blkg reference
289 * @blkg: blkg to get
290 *
291 * The caller should be holding an existing reference.
292 */
293 static inline void blkg_get(struct blkcg_gq *blkg)
294 {
295 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
296 atomic_inc(&blkg->refcnt);
297 }
298
299 void __blkg_release_rcu(struct rcu_head *rcu);
300
301 /**
302 * blkg_put - put a blkg reference
303 * @blkg: blkg to put
304 */
305 static inline void blkg_put(struct blkcg_gq *blkg)
306 {
307 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
308 if (atomic_dec_and_test(&blkg->refcnt))
309 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
310 }
311
312 struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
313 bool update_hint);
314
315 /**
316 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
317 * @d_blkg: loop cursor pointing to the current descendant
318 * @pos_css: used for iteration
319 * @p_blkg: target blkg to walk descendants of
320 *
321 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
322 * read locked. If called under either blkcg or queue lock, the iteration
323 * is guaranteed to include all and only online blkgs. The caller may
324 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
325 * @p_blkg is included in the iteration and the first node to be visited.
326 */
327 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
328 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
329 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
330 (p_blkg)->q, false)))
331
332 /**
333 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
334 * @d_blkg: loop cursor pointing to the current descendant
335 * @pos_css: used for iteration
336 * @p_blkg: target blkg to walk descendants of
337 *
338 * Similar to blkg_for_each_descendant_pre() but performs post-order
339 * traversal instead. Synchronization rules are the same. @p_blkg is
340 * included in the iteration and the last node to be visited.
341 */
342 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
343 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
344 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
345 (p_blkg)->q, false)))
346
347 /**
348 * blk_get_rl - get request_list to use
349 * @q: request_queue of interest
350 * @bio: bio which will be attached to the allocated request (may be %NULL)
351 *
352 * The caller wants to allocate a request from @q to use for @bio. Find
353 * the request_list to use and obtain a reference on it. Should be called
354 * under queue_lock. This function is guaranteed to return non-%NULL
355 * request_list.
356 */
357 static inline struct request_list *blk_get_rl(struct request_queue *q,
358 struct bio *bio)
359 {
360 struct blkcg *blkcg;
361 struct blkcg_gq *blkg;
362
363 rcu_read_lock();
364
365 blkcg = bio_blkcg(bio);
366
367 /* bypass blkg lookup and use @q->root_rl directly for root */
368 if (blkcg == &blkcg_root)
369 goto root_rl;
370
371 /*
372 * Try to use blkg->rl. blkg lookup may fail under memory pressure
373 * or if either the blkcg or queue is going away. Fall back to
374 * root_rl in such cases.
375 */
376 blkg = blkg_lookup_create(blkcg, q);
377 if (unlikely(IS_ERR(blkg)))
378 goto root_rl;
379
380 blkg_get(blkg);
381 rcu_read_unlock();
382 return &blkg->rl;
383 root_rl:
384 rcu_read_unlock();
385 return &q->root_rl;
386 }
387
388 /**
389 * blk_put_rl - put request_list
390 * @rl: request_list to put
391 *
392 * Put the reference acquired by blk_get_rl(). Should be called under
393 * queue_lock.
394 */
395 static inline void blk_put_rl(struct request_list *rl)
396 {
397 /* root_rl may not have blkg set */
398 if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
399 blkg_put(rl->blkg);
400 }
401
402 /**
403 * blk_rq_set_rl - associate a request with a request_list
404 * @rq: request of interest
405 * @rl: target request_list
406 *
407 * Associate @rq with @rl so that accounting and freeing can know the
408 * request_list @rq came from.
409 */
410 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
411 {
412 rq->rl = rl;
413 }
414
415 /**
416 * blk_rq_rl - return the request_list a request came from
417 * @rq: request of interest
418 *
419 * Return the request_list @rq is allocated from.
420 */
421 static inline struct request_list *blk_rq_rl(struct request *rq)
422 {
423 return rq->rl;
424 }
425
426 struct request_list *__blk_queue_next_rl(struct request_list *rl,
427 struct request_queue *q);
428 /**
429 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
430 *
431 * Should be used under queue_lock.
432 */
433 #define blk_queue_for_each_rl(rl, q) \
434 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
435
436 static inline void blkg_stat_init(struct blkg_stat *stat)
437 {
438 u64_stats_init(&stat->syncp);
439 }
440
441 /**
442 * blkg_stat_add - add a value to a blkg_stat
443 * @stat: target blkg_stat
444 * @val: value to add
445 *
446 * Add @val to @stat. The caller is responsible for synchronizing calls to
447 * this function.
448 */
449 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
450 {
451 u64_stats_update_begin(&stat->syncp);
452 stat->cnt += val;
453 u64_stats_update_end(&stat->syncp);
454 }
455
456 /**
457 * blkg_stat_read - read the current value of a blkg_stat
458 * @stat: blkg_stat to read
459 *
460 * Read the current value of @stat. This function can be called without
461 * synchroniztion and takes care of u64 atomicity.
462 */
463 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
464 {
465 unsigned int start;
466 uint64_t v;
467
468 do {
469 start = u64_stats_fetch_begin_irq(&stat->syncp);
470 v = stat->cnt;
471 } while (u64_stats_fetch_retry_irq(&stat->syncp, start));
472
473 return v;
474 }
475
476 /**
477 * blkg_stat_reset - reset a blkg_stat
478 * @stat: blkg_stat to reset
479 */
480 static inline void blkg_stat_reset(struct blkg_stat *stat)
481 {
482 stat->cnt = 0;
483 }
484
485 /**
486 * blkg_stat_merge - merge a blkg_stat into another
487 * @to: the destination blkg_stat
488 * @from: the source
489 *
490 * Add @from's count to @to.
491 */
492 static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
493 {
494 blkg_stat_add(to, blkg_stat_read(from));
495 }
496
497 static inline void blkg_rwstat_init(struct blkg_rwstat *rwstat)
498 {
499 u64_stats_init(&rwstat->syncp);
500 }
501
502 /**
503 * blkg_rwstat_add - add a value to a blkg_rwstat
504 * @rwstat: target blkg_rwstat
505 * @rw: mask of REQ_{WRITE|SYNC}
506 * @val: value to add
507 *
508 * Add @val to @rwstat. The counters are chosen according to @rw. The
509 * caller is responsible for synchronizing calls to this function.
510 */
511 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
512 int rw, uint64_t val)
513 {
514 u64_stats_update_begin(&rwstat->syncp);
515
516 if (rw & REQ_WRITE)
517 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
518 else
519 rwstat->cnt[BLKG_RWSTAT_READ] += val;
520 if (rw & REQ_SYNC)
521 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
522 else
523 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
524
525 u64_stats_update_end(&rwstat->syncp);
526 }
527
528 /**
529 * blkg_rwstat_read - read the current values of a blkg_rwstat
530 * @rwstat: blkg_rwstat to read
531 *
532 * Read the current snapshot of @rwstat and return it as the return value.
533 * This function can be called without synchronization and takes care of
534 * u64 atomicity.
535 */
536 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
537 {
538 unsigned int start;
539 struct blkg_rwstat tmp;
540
541 do {
542 start = u64_stats_fetch_begin_irq(&rwstat->syncp);
543 tmp = *rwstat;
544 } while (u64_stats_fetch_retry_irq(&rwstat->syncp, start));
545
546 return tmp;
547 }
548
549 /**
550 * blkg_rwstat_total - read the total count of a blkg_rwstat
551 * @rwstat: blkg_rwstat to read
552 *
553 * Return the total count of @rwstat regardless of the IO direction. This
554 * function can be called without synchronization and takes care of u64
555 * atomicity.
556 */
557 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
558 {
559 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
560
561 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
562 }
563
564 /**
565 * blkg_rwstat_reset - reset a blkg_rwstat
566 * @rwstat: blkg_rwstat to reset
567 */
568 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
569 {
570 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
571 }
572
573 /**
574 * blkg_rwstat_merge - merge a blkg_rwstat into another
575 * @to: the destination blkg_rwstat
576 * @from: the source
577 *
578 * Add @from's counts to @to.
579 */
580 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
581 struct blkg_rwstat *from)
582 {
583 struct blkg_rwstat v = blkg_rwstat_read(from);
584 int i;
585
586 u64_stats_update_begin(&to->syncp);
587 for (i = 0; i < BLKG_RWSTAT_NR; i++)
588 to->cnt[i] += v.cnt[i];
589 u64_stats_update_end(&to->syncp);
590 }
591
592 #else /* CONFIG_BLK_CGROUP */
593
594 struct blkcg {
595 };
596
597 struct blkg_policy_data {
598 };
599
600 struct blkcg_policy_data {
601 };
602
603 struct blkcg_gq {
604 };
605
606 struct blkcg_policy {
607 };
608
609 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
610
611 static inline struct cgroup_subsys_state *
612 task_get_blkcg_css(struct task_struct *task)
613 {
614 return NULL;
615 }
616
617 #ifdef CONFIG_BLOCK
618
619 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
620 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
621 static inline void blkcg_drain_queue(struct request_queue *q) { }
622 static inline void blkcg_exit_queue(struct request_queue *q) { }
623 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
624 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
625 static inline int blkcg_activate_policy(struct request_queue *q,
626 const struct blkcg_policy *pol) { return 0; }
627 static inline void blkcg_deactivate_policy(struct request_queue *q,
628 const struct blkcg_policy *pol) { }
629
630 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
631
632 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
633 struct blkcg_policy *pol) { return NULL; }
634 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
635 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
636 static inline void blkg_get(struct blkcg_gq *blkg) { }
637 static inline void blkg_put(struct blkcg_gq *blkg) { }
638
639 static inline struct request_list *blk_get_rl(struct request_queue *q,
640 struct bio *bio) { return &q->root_rl; }
641 static inline void blk_put_rl(struct request_list *rl) { }
642 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
643 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
644
645 #define blk_queue_for_each_rl(rl, q) \
646 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
647
648 #endif /* CONFIG_BLOCK */
649 #endif /* CONFIG_BLK_CGROUP */
650 #endif /* _BLK_CGROUP_H */
This page took 0.060774 seconds and 6 git commands to generate.